[c-nsp] ASR 9000 Series MPLS Label Limits

James Bensley jwbensley at gmail.com
Mon Jan 30 10:13:38 EST 2017


Hi All,

With the recent post about ASR920 MPSL label limits that has reminded
me to revisit this thread.

- Does anyone know exactly what that TCAM label limit is for ASR9K's
(Typhoon and Tomahawk and different RSPs etc), more specifically how
it split between ingress and egress labels?

- What commands can be used to show the usage?

Lukas provided a link to BRKSPG-2904 which states a 1M label limit for
Typhoon cards and since the label space is 2^20 that would accommodate
all labels.

In some platforms the label space is split into egress and ingress
labels. More specifically I mean labels received over LDP and BGP for
a prefix/VRF and the local device is the tail end LER of an LSP. The
local device may allocation a local label which is a separate label
for it’s local prefix inside that L3 VPN (for example) and then
advertise that local label on (or not). Also the local device might
receive a label and then allocate another for advertising on with no
local access interface (MPLS Option B for example).

In the case of ME3600X/ME3800X there is one shared label space in the
NILE TCAM, not all of that space can be used for local labels though
so it can hold more egress labels than it can ingress. In addition to
this the problem is exacerbated by the use of MPLS Option B links
which double up label allocations for prefixes sent over the Option B
link.

I have made some notes here specifically for the ME3600X/ME3800X:
https://null.53bits.co.uk/index.php?page=inter-as-mpsl-optionb-mpls-label-usage

Does anyone know if the Trident/Typhoo/Tomahawk TCAMs are split into
locally allocated labels and received labels, or is it just one shared
space of 1M labels? If it is one shared space and the device receives
500K labels and allocates local labels and then advertise them on,
that will be 1M labels used but for only 500k entries? Also if the
MPLS Option B behaviour is the same double allocating as with the ME’s
occurring meaning  it can only receive ~333k labels, ~333k are locally
allocated and another ~333k for an Option B?

Outputs below from an example ASR9001;

Cheers,
James.



RP/0/RSP0/CPU0:9001#show mpls forwarding summary
Forwarding entries:
   Label switching: 128860, protected: 21


RP/0/RSP0/CPU0:9001#show bgp label summary
----------------  ------
TOTAL             128860


RP/0/RSP0/CPU0:9001#show controllers np struct Label-UFIB summary
                Node: 0/0/CPU0:
----------------------------------------------------------------
NP: 0  Struct 28: MPLS_LEAF
Struct is a PHYSICAL entity
Reserved Entries: 0, Used Entries: 224542, Max Entries: 2097152

NP: 1  Struct 28: MPLS_LEAF
Struct is a PHYSICAL entity
Reserved Entries: 0, Used Entries: 224543, Max Entries: 2097152

! ^ Does this mean that we can store 2M labels? So we can store 1M
ingress and 1M egress? Or is this something completely different?


More information about the cisco-nsp mailing list